1: \begin{abstract}

2: In this paper we study the performance of the Projected Gradient Descent

3: (PGD) algorithm for $\ell_{p}$-constrained least squares problems

4: that arise in the framework of Compressed Sensing. Relying on the

5: Restricted Isometry Property, we provide convergence guarantees for

6: this algorithm for the entire range of $0\leq p\leq1$, that include

7: and generalize the existing results for the Iterative Hard Thresholding

8: algorithm and provide a new accuracy guarantee for the Iterative Soft

9: Thresholding algorithm as special cases. Our results suggest that

10: in this group of algorithms, as $p$ increases from zero to one, conditions

11: required to guarantee accuracy become stricter and robustness to noise

12: deteriorates.\end{abstract}

13: